How an SME Picks Its First AI Use Case (Without Boiling the Ocean)

If you have been "looking at AI" for six months and still have no project starting on Monday, the problem is not the shortlist. It is the size of the question.

In 2026 AI adoption is no longer under any doubt. AI will either enter through governed plans sanctioned by the company, or through shadow-AI. The challenge facing companies now is what should AI be used for?

Paralysis By Analysis

Also known as "boiling the ocean" — companies risk following the impulse to evaluate every possible AI use case, build a heat map, score them, prioritise them, and then never start any of them because the prioritisation never converges. While adopting AI should be done in a controlled manner, it requires a willingness to disrupt the whirlwind of day-to-day business and there is never enough time as it is. An SME does not have a six-month strategy team. The CFO is doing payroll. The COO is in a customer call. The leadership group has Tuesday morning.

So here is the version that works in Tuesday-morning blocks.

• Four criteria
• one shortlist
• one project
• one quarter.

The four criteria for a first AI use case

The job of the first AI project is not to be the highest-impact thing the business could do with AI. It is to finish, to produce evidence, and to teach the organisation what its actual AI operating model looks like. Once you have one finished project, the second one chooses itself. The hard one is the first.

A first AI use case should be:

1. Bounded in scope. One team, one workflow, simple inputs, one output. If two teams or two workflows are involved, you are picking your second project, not your first.
2. Measurable in outcome. Measuring is not easy, but is vital for confidence. Ideally, before the project starts, write down your metrics — perhaps time saved per task, an error rate reduction, or throughput per person. It is important to include baselines where known, otherwise success is something you will discover over time. If you have no idea, identify some aims - what does success look like?
3. Reversible in commitment. The tooling you are putting in place is not architecturally load-bearing. If the pilot fails, you can stop using the tool and the rest of the business carries on. No data migration. No process redesign. No 18-month contract.
4. Owned by a single person. Not a committee. Not "the AI working group." A named person who is on the hook for the outcome and who has authority to make the day-to-day decisions about the pilot. If three people own it, nobody owns it. The rest of the company or team agrees to support the owner to make the project a success.

Take these criteria seriously - build to succeed!

You may find a lot of your candidate projects fail at least one of these, but AI needs to be proved to be useful, then adoption can expand. The projects that pass all four criteria are often unglamorous, proposal first-drafts, contract review summaries, internal knowledge search, sales call transcript summaries, customer support ticket triage.

None of those will impress at a board meeting but you should be able to deliver results in under 90 days.

Where this fits in SAFE-AI

This is the operational meat of [[Stage 1 - Assess]] feeding into [[Stage 2 - Strategise]]. The Assess stage establishes what AI is already in the building, what data you have to work with, and what your readiness is across people, process, and governance. The Strategise stage turns that picture into a prioritised plan with a first project named. The four criteria above are how to pick the first project from your candidate list.

In practice, the conversation goes like this. SAFE-AI Assess gives you a list of 8 to 15 candidate use cases — drawn from interviews, from the existing process map, from the shadow AI baseline, from existing company strategy. SAFE-AI Strategise scores candidates against the four criteria. The first candidate to clear all four becomes the pilot. The runner-up gets parked for project two.

Crucially, "highest impact" is not in the criteria. That is intentional. The highest-impact project is usually the most complex one — the one that touches multiple teams and depends on data that is not yet clean and requires a vendor decision that the board has to ratify. That project is not your first project. It might be your fourth.

The "first project veto list"

The five things to NOT pick as a first AI project:

• Customer-facing AI without a fallback. Chatbots, customer support automation, anything where the customer talks to the AI directly. The risk profile is wrong for a first project — failures are visible to customers and the rollback hurts.
• Anything requiring data cleaning as a precondition. If the project starts with "first we need to clean up the customer database," your first project is the data-cleaning project, not the AI project. Sequence matters.
• Anything that depends on a vendor decision that has not been made. "We will pick the AI platform later" is not a first-project plan. Pick the platform first as a separate decision; the project then runs on the platform.
• Anything where the success metric has to be invented. If the team has never measured the thing the AI is meant to improve, the project will spend its first six weeks measuring the baseline and never get to the AI.
• Strategic transformation projects. "We are using AI to reposition the business" is a multi-year programme. The first project is a 90-day proof.

What to do with this

If you have a candidate AI project and you are not sure whether it clears the four criteria, the AI Readiness Scorecard will give you a short structured read on where your organisation actually is — against governance, data readiness, and operating model — in about five minutes.

If you have a shortlist and you want to to explore how the Koallabs SAFE-AI framework can help you succeed, book a 30-minute discovery call. Bring the shortlist and we can explore it together.

The point is to start. The first project is the hard one only because it is unstarted. Once it is finished — even if it is the unglamorous one — the second project picks itself.

Want to see where your organisation stands? Take the AI Readiness Scorecard — 6 questions, personalised AI report.

Ready to scope an engagement? Book a 30-minute discovery call — no sales script, just a conversation.